Quanta SDK Empirical Benchmark Results (v1.2.0-production)
Benchmark Date: September 2026
Test Platform: Apple Silicon M-Series (macOS Darwin ARM64), Python 3.12+, NumPy 2.x, PyTorch 2.x, MLX
Verification Suite: benchmarks/run_paper_benchmarks.py · 2,076 Passing Automated Tests
Software DOI: GitHub Release v1.2.0
1. Simulation & Hardware Acceleration Benchmarks
| Benchmark Scenario |
Scale / Configuration |
Execution Time |
Throughput / Speedup |
Notes & Verifications |
| SIMD Clifford Engine |
100,000 Random Clifford Gates |
31.91 ms |
3,133,813 gates/s |
Aaronson-Gottesman binary tableau |
| Apple Silicon MLX (Metal) |
24-Qubit Dense Statevector |
78.42 ms |
52.09× Peak Speedup |
Zero-copy Unified Memory (vs CPU 4.08s) |
| Matrix Product States (MPS) |
250-Qubit GHZ State Preparation |
3.42 ms |
Low-rank tensor network |
Schmidt truncation \(\chi=64\) |
| Statevector Bell State |
2 Qubits, 1024 Shots |
0.14 ms |
End-to-end pipeline |
Build + DAG + sim + measurement |
| Statevector GHZ State |
10 Qubits (\(2^{10}\) amplitudes) |
0.28 ms |
Exact statevector |
\(\|\|\psi\|\|_2 = 1.000000\) |
| Statevector GHZ State |
20 Qubits (\(2^{20}\) amplitudes) |
64.12 ms |
Dense tensor contraction |
1,048,576 complex128 amplitudes |
| Grover Search |
8 Qubits, Target=42 |
0.18 ms |
Quadratic oracle speedup |
\(P(\text{target}) > 0.99\) |
| VQE Ground State (\(H_2\)) |
2 Qubits, 2-Layer Ansatz |
184.50 ms |
Exact convergence |
3-term Pauli decomposition |
2. Quantum Error Correction (FTQC) Decoding Latencies
Empirical execution times for syndrome extraction and decoding:
| QEC Decoder & Code |
Architecture / Parameters |
Decode Time (ms) |
Success / Fidelity |
| Gross qLDPC Normalized BP-OSD-0 |
Canonical \([[144, 12, 12]]\) Bivariate Bicycle |
1.54 ms |
12 Logical Qubits, \(12\times\) qubit saving |
| Edmonds Blossom MWPM |
Rotated Surface Code \(d=3\) |
0.29 ms |
Exact minimum weight matching |
| Edmonds Blossom MWPM |
Rotated Surface Code \(d=5\) |
0.82 ms |
Sub-millisecond FTQC loop |
| Edmonds Blossom MWPM |
Rotated Surface Code \(d=7\) |
2.41 ms |
Full spacetime graph matching |
| Steane Code Lookup Decoder |
Steane \([[7, 1, 3]]\) |
0.04 ms |
Single-qubit error correction |
| 15-to-1 Bravyi-Kitaev Distillation |
Magic State $ |
T\rangle$ Factory |
0.62 ms |
3. Autograd & Mathematical Gradient Benchmarks (quanta.torch)
| Gradient Method |
Mathematical Precision |
Latency |
Baseline Error |
| Daleckii-Krein Fréchet Autograd |
Machine Precision (\(\epsilon = 9.99\times 10^{-16}\)) |
0.12 ms |
Padé approximation error: \(>1.3\times 10^{-6}\) |
| Parameter-Shift Rule (Exact) |
Analytically Exact (\(\epsilon = 0\)) |
0.38 ms |
5 circuit evaluations (\(2n+1\)) |
| Natural Gradient (QFIM) |
Fubini-Study Riemannian metric |
0.84 ms |
Exact Fisher information inversion |
Methodology & Reproducibility
- Zero Mock Policy: All benchmark figures are measured on actual silicon execution paths; no synthetic sleeps or interpolated values are permitted.
- Timing Protocol: Median of 5 consecutive runs with garbage collection suppression.
- Execution Script: Run
python benchmarks/run_paper_benchmarks.py to reproduce the full benchmark matrix.
Authorship & Attribution
- Principal Architect & Author: Abdullah Enes SARI (ORCID: 0000-0002-8827-0587)
- Affiliation: ONMARTECH Quantum Computing Initiative (
info@onmartech.com)